Merge model types for Hathor

[GunnarAtEgde]

Hathor is currently using multiple models for the same things. We need to clean up this code.

Fixes needed:

1. There are two identical copies of the vehicleTypeTypes.ts file, one in packages\vtype-details and the other in data\vehicle-types. There is mention of merging the packages\vtype-details into Hathor, so remember then to delete the extra copy.

2. Vehicle in data\vehicles\types and in data\vehice-types\vehicleTypeTypes.ts

3. VehicleGQLShaped and VehicleWire. Issues here:

• *Wire have netexId, the other have id, but other than that they are very similar
• Note that a lot of properties need to be "optional", as they are sometimes used and sometimes not
• If one finds reason to keep more than one type, at least the copy-ing between them needs to made generic. Currently, there is great risk of properties being "missed" because either the fetch or update functions don't map them

[dynnamitt]

Concrete instance of item 3 — dup to be fixed in parallel.

From the #101 review: fetchVehicle is a near-verbatim copy of fetchVehicles' projection map — data/vehicles/api/fetchVehicles.ts:52-70 vs :95-113. This is exactly the "copying between them needs to be made generic … great risk of properties being missed" concern: the two already show drift (the copied truncation warning still logs fetchVehicles:).

Fix alongside the type merge — extract a single toVehicleGQLShaped(v: VehicleWire): VehicleGQLShaped (+ shared truncation-warn helper) and reuse from both fetchers. Tracked inline on #101.

[dynnamitt]

version type drift — another instance of item 3 (from the #101 review).

The SDL types version: String (on both Vehicle and VehicleType) and Sobek returns it as a string ("1"/"3", round-trip-confirmed), but VehicleWire and VehicleGQLShaped both declare it number and the projection passes it through untouched — so the runtime value is a string under a number type. Classic "copy between Wire↔Shaped silently mishandles a property."

• Wire (lies): data/vehicles/api/fetchVehicles.ts:14, nested :21
• Passthrough (no coercion): :54 / :97 (the dup), nested :65 / :108
• Shaped: data/vehicles/types/vehicleGqlShaped.ts:11, :21
• Numeric consumer: data/vehicles/utils/vehicleSortValue.ts:25

Suggested fix — coerce at the seam. Make the wire type tell the truth (string, matching the SDL) and convert to number once in the projection; VehicleGQLShaped then stays honestly numeric and the sort comparator is untouched.

// VehicleWire (+ nested transportType)
version?: string;

// projection
version: v.version != null ? Number(v.version) : undefined,

This folds straight into the toVehicleGQLShaped extraction noted above — do it once there instead of at both :54 and :97.

(Alternative: type version as string end-to-end and compare via Number() in the sorter — more churn, only needed if versions can be non-numeric.)

[dynnamitt]

Target the graphql-codegen shape

Ran real @graphql-codegen/cli (typescript + typescript-operations) against src/graphql/sobek.schema.graphqls + the VehicleType query/mutation to derive the canonical vocabulary. The cleanup below = converge on that shape and relocate the model layer out of features (it's all shared).

Current muddy state — vehicle-types vs vehicles

Layer	vehicle-types	vehicles
result (output)	VehicleTypeWire (api/)	VehicleWire (api/)
project (wire→domain)	projectVehicleType() — named, 1×	inline mapper, duplicated ×2 (fetchVehicles+fetchVehicle)
domain (id)	VehicleType ✓ clean	VehicleGQLShaped ✗ leaky name
serialize (domain→input)	serializeVehicleType() full-document, in api/	none — built inline in the hook (useVehiclePairSave, orUndef omit-blank)
input (netexId)	VehicleTypeInput	VehicleInput

Two parallel-but-divergent hand-rolled stacks; different names, placement (api vs hook), and blank-policy (full-doc vs omit-blank — the source of the destructive-save risk). No <Entity>Output type exists — the query result, hand-typed as *Wire, plays that role.

Side-by-side: codegen canonical vs current (extended)

codegen (canonical)	current hathor	location	target home
VehicleType (output) / VehicleTypesQuery result	VehicleTypeWire	data/vehicle-types/api/fetchVehicleTypes.ts	shared (generated)
VehicleTypeInput	VehicleTypeInput ✓ 1:1	data/vehicle-types/api/	shared (generated)
(no codegen equiv — app domain)	VehicleType (id)	data/vehicle-types/types/vehicleTypeTypes.ts	feature vehicle-types
MultilingualString(Input)	Name	data/vehicle-types/types/ (imported by vehicles + deck-plans)	shared
DeckPlan(Input)	DeckPlan	data/vehicle-types/types/ (deck-plans reaches in)	shared
Vehicle (nested)	Vehicle	data/vehicle-types/types/ (the vehicles entity, wrong feature)	shared
TransportMode (21)	TransportMode (13-subset + 'unknown')	data/netex/transportMode.ts	shared (see #93)
PropulsionType/FuelType/HybridCategory	same names	data/vehicle-types/types/	shared
PassengerCapacity(Input)	PassengerCapacity (+ Input hand-rolled)	data/vehicle-types/types/ + api/	shared
FareClass	— missing —	—	shared
PrivateCode(Input) / KeyValues(Input)	— missing — (dropped on save)	—	shared
Page / VehicleTypePage	Page<T> ✓	graphql/paginationTypes.ts	shared ✓ (already)
VehicleTypeFilter	inline on VehicleTypeVars (stale: missing name)	graphql/queries/fetchVehicleTypes.ts	shared (generated)
VehicleTypesQueryVariables	VehicleTypeVars	graphql/queries/	feature (op vars)
MutateVehicleTypeMutation	CreateOrUpdateVehicleTypeResponse	graphql/mutations/	feature (op result)

Placement findings (beyond types — same layer+sharing rule)

Thing	Layer	current location	target home
NetexId (+stories)	shared component	data/netex/NetexId.tsx	components/ (next to TransportModeIcon)
transportMode model	shared model	data/netex/transportMode.ts	shared model layer
transportModeFilters (chips+labelKeys)	presentation	fused into the model file	UI side (see #93)
iconLoaderUtils	theming util, 1 consumer (HeaderActions)	src/utils/	theming subsystem (theme/ or components/icons/)
compareWithEmptyLast	sort primitive + a Sobek #121 whitespace trim	src/utils/	utils ✓ (move the #121 trim to model-normalization)
commitSave	generic async primitive	src/utils/	utils ✓

Lift verdict

A codegen module is inherently cross-feature (one file holds Vehicle, VehicleType, DeckPlan, Organisation + every value object) → the entire backend-model layer is shared. ~10 types hathor keeps inside data/vehicle-types/ are reached into by vehicles/deck-plans — the coupling smell. Only the operation documents + a deliberate domain projection are feature-owned.

src/
  lib/graphql/generated.ts   # whole schema: types/inputs/enums/ops — SHARED
  components/netex/NetexId.tsx
  features/vehicle-types/{api(op docs), model(project/serialize), components}
  features/vehicles/ …       # imports Vehicle/Name/etc. from lib, not from vehicle-types

Already landed (PR #114, partial progress on item 3)

• <Entity>Input convention added on both sides (VehicleTypeInput/VehicleInput) — matches codegen 1:1; dropped the as VehicleWire cast.
• description + out-of-set transportMode now round-trip (were silently nulled). FareClass/PrivateCode/KeyValues still unmodelled.

Proposed scope for this issue

Widen #107 from "merge model types" → "merge + relocate the model layer onto the codegen shape, and place shared code by layer." TransportMode's UI/enum specifics tracked in #93. (Repro: graphql-codegen with typescript+typescript-operations over the SDL + op docs.)

[dynnamitt]

Note — this reverses the "no src/types/" stance (needs a superseding ADR)

The fork's recorded design deliberately excluded a global shared model/types dir:

• OPEN_QUESTIONS.md → "Hathor does not currently have a src/types/ dir — keep it that way" (cross-feature NeTEx types → src/data/netex/; other shared consts → flat file at the LCA, e.g. src/data/graphqlErrMsg.ts).
• FORK_DECISIONS ADR 2026-05-28 → bulletproof-react segmentation is "per-feature, not global."

Adopting the codegen shape reverses that — it needs a shared model home (src/lib/graphql/generated.ts or a types/). When this lands, record a superseding ADR in FORK_DECISIONS.md.

Rationale that overrides the old call (so it reads as intentional, not a flip-flop): the prior objection was "a types/ dir is an orphan-junk magnet" — true for hand-rolled shared types, but a generated, schema-synced, CI-drift-guarded module is the single source of truth, the opposite of orphan junk. The objection doesn't apply to codegen output. (Memory note feedback_avoid_orphan_types_dir flips with it.)